Namelix

@rhysbeingsocial yeah sure - first it passes your input keywords through fasttext - we get a random sampling of the top 100 or so words most related to your input keywords - the top words are used as the dictionary input to a markov chain word generator, which generates 10~15000 random words - the results are passed again through fasttext, we take the cosine distance of the resulting word to your input keywords (by using the out-of-vocab feature of fasttext) - the top 100 most relevant results are returned I had previously tried an encoder-decoder model with fasttext as the encoder and an RNN as the decoder, but it didn't work well so I ended up with this approach.

@rhysbeingsocial @jak_9994 that just sounds like an algorithm. Not AI.

@rhysbeingsocial @mickc79 AI is just algorithms.. word vectors, neural nets and earlier NLP research all qualify as AI imo

@rhysbeingsocial @jak_9994 well I could implement what you just described. It’s simply a step by step process that’s repeated every time. Not an AI. However in a world where “AI” is thrown into almost every description of every new product, I don’t blame you.

@itsthisjustin it gives it some flavor I guess : ] each name is rendered as a logo from my logo maker site. The colors are randomly generated from a list, I'm removing the illegible ones as I find them.

@jak_9994 I'd recommend using HSL color values instead of hex RGB, and making sure there is at least a 50% gap in the Lightness channel between the background and foreground colors. That should help prevent legibility issues.